Content uploaded by Luke K Fryer
Author content
All content in this area was uploaded by Luke K Fryer on Jun 06, 2020
Content may be subject to copyright.
Language Learning & Technology
ISSN 1094-3501
June 2020, Volume 24, Issue 2
pp. 8–22
EMERGING TECHNOLOGIES
Copyright © 2020 Luke K. Fryer, David Coniam, Rollo Carpenter, & Diana Lăpușneanu
Bots for language learning now: Current and future
directions
Luke K. Fryer, University of Hong Kong, Faculty of Education (CETL), Hong Kong
David Coniam, The Education University of Hong Kong, Hong Kong
Rollo Carpenter, Cleverbot, UK
Diana Lăpușneanu, Mondly, România
Abstract
Bots are destined to dominate how humans interact with the internet of things that continues to
grow around them. Despite their still budding intellectual capacity, major companies (e.g.,
Apple, Google and Amazon) have already placed (chat)bots at the centre of their flagship
devices. (Chat)Bots currently fill the internet acting as guides, merchants and assistants.
Chatbots, designed as communicators, however, have yet to make a meaningful contribution to
perhaps their most natural vocation: foreign language learning partners. This review engages in
three questions that surround this issue:
1. Why are chatbots not already at the centre of foreign language learning?
2. What are two key developers of chatbots working towards that might push chatbots into
the language learning spotlight?
3. What might researchers, educators, and developers together do to support chatbots as
foreign language learning partners right now?
Keywords: Bots, Chatbots, Conversational Agents, Language Learning
Language(s) Learned in This Study: All spoken languages are addressed
APA Citation: Fryer, L. K., Coniam, D., Carpenter, R., & Lăpușneanu, D. (2020). Bots for language
learning now: Current and future directions. Language Learning & Technology, 24(2), 8–22. Retrieved
from http://hdl.handle.net/10125/44719
Introduction
(Chat)Bots are now in a position to radically change how we interact with our growing digital
world (Dale, 2016), from reading and writing to listening and speaking. One of the many
revolutions that chatbots will kick off is how (and in many cases whether) we learn a new
language. In this review, it is argued that chatbots will eventually be the perfect language-
learning partner, potentially enabling us to learn multiple languages anywhere, anytime and at
our own pace.
Despite this eventual reality being clearly on the horizon for more than a decade (Fryer &
Carpenter, 2006; Coniam, 2004), there is little direct evidence to indicate that a golden age in
language learning opportunities is upon us. Other areas where chatbots are destined to dominate,
however, have been gathering substantial momentum. Simple internet searches show that
adjusting one’s home environments (i.e., heat, light, etc.), and navigating one’s media (music,
movies, etc.) are all becoming seamless interactions with one of a handful of established
Luke K. Fryer, David Coniam, Rollo Carpenter, and Diana Lăpușneanu.
9
conversational agents (chatbots) such as Google, Siri or Alexa. In addition to the ubiquitous use
of chatbots within e-commerce and website support, chatbots have also made inroads into
increasingly specific areas such as formal (Cameron et al., 2017) and informal counselling on
sites like Replika, specifically in areas such as smoking cessation (Dubosson, Schaer, Savioz, &
Schumacher, 2017); broader health care issues (Shah & Philip, 2019); supporting students in
course choices (Fleming, Riveros, Reidsema, & Achilles, 2018); educating sensitive populations
about sex, drugs, and alcohol (Crutzen, Peters, Portugal, Fisser, & Grolleman, 2011); as well, of
course, as personal assistants (Daniel, Matera, Zaccaria, & Dell’Orto, 2018). And yet, in spite of
these advances, chatbots are still, at best, weak, supplemental language learning partners (Fryer,
Nakao, & Thompson, 2019).
The present prospective review begins by discussing what has changed over the past decade: the
user and the interfacing technology supporting interaction. The review will then proceed to
discuss two cutting-edge chatbots from the developers’ own perspectives: one, Cleverbot, with
more than two decades of developmental history and another, Mondly, which was founded in
2014 and designed specifically to support language learners. These chatbots will be discussed
with a focus on their current usefulness in the area of language learning and where they are
heading in the next few years. The final component of this prospective review will discuss
directions forward. It acknowledges the fact that the technology will continue to grow. However,
it notes that we need to start using the tools we have—and, in some cases, fully developed
chatbots we already have—rather than waiting for a major breakthrough in AI communication
thinking and language skills. In this section, research in the area of multimedia learning generally
and chatbot interaction specifically will be discussed, finally indicating how we might get started
and steadily build on the still-developing tools at hand.
Where We Are
Users and Technology
Where Eliza’s (Weizenbaum, 1966) early users might have found communicating through
typed text novel, today’s users are thoroughly primed to communicate with an online entity
(Jiang, 2018). During the past three decades, communicating online has gone from the fringe to
the norm. To suggest that the present generation is comfortable with online synchronous and
asynchronous communication via text or speech, is an understatement (Vogels, 2019).
Chatbots are already readily available online within most messaging applications and many
information-orientated websites (e.g., universities, libraries, and museums). The tremendous
popularity of two recent chatbots, one in Mainland China XiaoIce (Zhou, Gao, Li, & Shum,
2018) and one in California, Facebook’s M (Simonite, 2017) have removed any questions
regarding readiness on the part of users to use and trust a chatbot.
During the past five years there has been huge growth in the language learning online software
sector with a host of language learning applications, such as Duolingo and Mondly, joining
more traditional applications like Rosetta Stone in recruiting huge user populations. This
suggests that despite the frequent heralding of a future where machine translation will make
learning a new language antiquated, there is still a strong appetite for learning languages.
However, it is clear that it is not potential users who are impeding progress toward a chatbot
language learning future. It must therefore be weaknesses in the technology itself. Identifying
10
Language Learning & Technology
these weaknesses and how they might be overcome, or at least ameliorated, is the natural next
step toward a chatbot language learning future.
Working our way back from the advances that have been made during the past decade, it is
important to highlight significant and still rapidly improving speech recognition software.
Speech recognition software has made it possible to make everything from short dictations to a
broad array of commands and requests to conversational agents. Most online agents have both
input and output speech functions, which has enabled the agents to move from being solely
text to aural communicants—and with relatively high reliability. Speech recognition and the
increasing ubiquity and power of mobile devices are key factors supporting chatbots’ use as
language-learning partners. Despite these apparent affordances, and the plethora of ways in
which speech recognition is already changing how we engage with media (Howell, 2019),
embarking on the challenging path of chatbot-centred language acquisition is not a particularly
popular venture. This is demonstrated by the scant number of companies building their
language learning teaching approaches firmly around this budding area of artificial intelligence
(see Mondly and Duolingo for two interesting examples of companies breaking ground in this
field).
Chatbots and Conversational Agents
Chatbots have been around for decades. The idea of chatbots as language learning partners is
more recent, but is slowly gathering momentum. As far back as the early 2000’s, Coniam (2004)
reviewed two chatbots with potential as language learning partners. The first was the ALICE
Artificial Intelligence Foundation site’s Dave, which they claimed to be the “perfect private
tutor,” since he replied “in perfect English, just like a private English teacher.” (Coniam, 2004, p.
160). Attempts at conversing with Dave indicated, however, that while many of Dave’s
conversational strategies had quite a natural feel about them, there were syntactic infelicities and
conversational glitches which indicated that the program was unlikely to pass the Turing Test in
the near future.
The second chatbot reviewed by Coniam was Lucy chatbot (now defunct). This chatbot had
support for beginners, where incorrect input by L2 learners was manually corrected, so that Lucy
was at times able to suggest corrections to certain grammatical errors. The online Lucy chatbot
was developed into a standalone piece of software, Lucy’s World: Smallt@lk. This reworking
utilised the speech and interactive elements of the Lucy chatbot, but controlled the situation in
that topics and situations that Lucy was able to converse about were restricted. Since these early
steps toward chatbots supporting language learning, both potential users and chatbots have
substantially changed. Users’ willingness to engage in online communication and language
learning have become commonplace. Yet, despite the fact that the tools necessary for easy aural
communication with digital agents are quickly becoming normal, chatbots are still not
dominating or even meaningfully infiltrating how we learn languages. To a reader considering
this idea for the first time and having little experience conversing with chatbots, the reason for
this gap might not be apparent. Millions of people have spoken with one of the major
conversational agents (Alexa, Siri or Google) and perhaps been surprised at their ability to be
helpful with straightforward tasks. If these users have ever tried to push the boundaries, however,
they know that, despite the name conversational agent or chatbot, these digital personas can
rarely get past a conversation that includes more than two exchanges. The fact is that despite
advances in Natural Language Processing (Hirschberg & Manning, 2015) and Deep Processing
(Kriegeskorte, 2015), each considered to be crucial to advancing AI technology, major
Luke K. Fryer, David Coniam, Rollo Carpenter, and Diana Lăpușneanu.
11
breakthroughs in chatbots’ communication skills have not been forthcoming (Dale, 2016). This
has pushed researchers to work towards compensating for AI gaps (Meszaros, Chandarana,
Trujillo, & Allen, 2017).
With these issues in mind, it is worth briefly reviewing two chatbots and their developers:
Cleverbot and Mondly’s chatbots. These two chatbots are from two ends of our current spectrum
of free-to-use chatbots. Cleverbot has a considerable history and has had success in competitions
(e.g., winning the Lobner Prize) and with users (it has had millions of exchanges with interested
online users). Cleverbot was not designed as a language-learning tool, but has seen considerable
incidental use as a language learning partner. The second chatbot is brand new and was
specifically designed from the ground up to support language learning. The Mondly chatbot was
developed to make the most of the affordances of cutting-edge technologies such as Augmented
and Virtual Reality (AR and VR).
Two Chatbot Developers, Where They Are and Where They Are Going
Cleverbot
As far back as 1988, on a tiny computer with just 1K of RAM, Rollo Carpenter created a
program that was able to talk back. In 1996—even before the founding of Google—Cleverbot’s
predecessor Jabberwacky went online and started talking to and learning from visitors. Over the
past 20 years, and with no marketing, millions of people have been talking to
Jabberwacky/Cleverbot.
Like many of the chatbots that have followed in its wake, Cleverbot and its precursors learn from
their users’ language. To date, Cleverbot has had more than 10 billion user communications, on
nearly every imaginable topic. If it has not experienced a subject or a context, it will not initially
know how to handle it. In this, it is a little like a person interacting socially, who is constantly
adapting. Cleverbot was not designed to be useful, but rather to keep people company and to
entertain. Cleverbot knows language only in the form of text-based communication. It has never
learned anything directly from the world, or from its creators; only from the responses given by
its users. It was not designed to support second or foreign language learning, but its abilities to
engage and amuse users in a language and on a subject more or less of their choosing, has made
it an unintentional digital island for human language learning. Like all chatbots, Cleverbot’s
strength comes in part from the fact that it is always there, and always replies. Cleverbot may
argue with its users, but it is also infinitely patient and non-judgmental. Reviewing its endless
stream of interactions with people the world over suggests that there are far fewer social
constraints when talking to a machine than is the case with human-human interaction (e.g., see
conversation extracts). It is filtered automatically to ensure it does not repeat the inappropriate
language some of its users communicate, but from the perspective of a person learning a
language, freedom to communicate at any time, without restraint or social pressure makes it
closer to the ideal conversational partner.
Cleverbot has its own brand of Artificial Intelligence software, a key concept of which is
context. While the Cleverbot AI outputs sentences said to it verbatim, the manner in which it
selects what to say is complex. It considers the last 50 interactions of the current conversation,
where available, comparing these non-precisely to millions of past conversations, looking for
equivalent contexts, or the optimal summation of small clues as to what might be the most
conversationally appropriate next response in the conversation. In January 2019, Cleverbot’s
pool of potential responses passed 500 million, yet it is still the case that almost every
12
Language Learning & Technology
conversation is unique. Approximately 50% of its data is in English, the remainder mostly
representing a range of other European languages.
Carpenter and colleagues at Existor are working on a new version of Cleverbot that will
understand the similarity of words, phrases, sentences and wider intentions at a deeper level,
entirely independently of the letters of the words themselves, and that will construct outputs
more in the manner that humans compose sentences (Carpenter, 2016). While most chatbots
currently available have specific purposes, are limited in scope, and are not able to learn,
Cleverbot aims to be general, to be a different conversational partner to each user, and to
genuinely engage users in conversation.
Using Existor’s tools or others’ tools, it is technically possible to purpose-build a bot for
language teaching, talking people through known learning sequences, correcting errors and more.
The problem with such an approach is that no company—even using the latest Machine Learning
techniques—has actually resolved the language problem: The ability of the machine to truly
understand. If the machine cannot truly understand, such language teaching modules have to be
constructed, or programmed, by hand—a never-ending task that would remain forever
incomplete. With billions of user inputs to work with, Carpenter and colleagues are well-placed
to create new tools and new machine language understanding. In the meantime, people around
the world already do talk at length to Cleverbot for language practice, and are incentivised to do
so by its engaging, humorous and always-available presence.
Mondly
In contrast to Cleverbot, whose history stretches back to before chatbots were widely
recognised as the future of the internet, the Mondly chatbot is only four years old. Mondly
chatbots—like an increasing number of such chatbots—was developed as part of a language-
learning platform, rather than as a means of entertaining human users as was
Jabberwacky/Cleverbot. In step with today’s users, it operates as smartphone-centred software,
and has been downloaded by 40 million users in over 190 countries. When Mondly was
launched in 2016, it was the first of its kind in the language learning niche. One year later,
Mondly began implementing chatbots with other new technologies (VR and AR]) as well.
From the very beginning, the Mondly chatbot was designed to interact with users, understand
voice input, and reply with a human voice. The chatbot’s goal was (and still is) to be engaging
and fun. However, at the same time, the Mondly chatbot aimed to provide adaptive lessons
(i.e., encourage learning through play; Smith & Pellegrini, 2008) that encourage users to
practice the language they are learning in everyday scenarios—such as ordering in a restaurant.
Mondly recognizes millions of inputs and creates an adaptive visual response when it
recognizes a word or phrase that the user has said, providing feedback that can support users’
confidence.
The initial Mondly chatbot software expanded to include 27 new languages (with six initial
languages and a final total of 33). Apart from personalized scripts for all its languages,
Mondly’s most crucial advance has been the development of a key phrase database. This
database represents the chatbots’ brains. The focus of ongoing work now is machine learning,
with the company’s aim being to make Mondly’s chatbot appear smarter and feel more like an
autonomous conversant. Adaptive learning is another frontier Mondly is seeking to push;
chatbots need to be able to create specific learning patterns for groups of learners that share
certain traits. For example, if a user tends to forget a word—how to say shark in Spanish for
Luke K. Fryer, David Coniam, Rollo Carpenter, and Diana Lăpușneanu.
13
example—the chatbot needs to know that it should ask the user that word more often to help
them retain the information.
Digital Assistants (such as Google, Siri and Alexa) are slowly enabling humans to control and
communicate with our intelligent houses, cars, and the wider world that surrounds us. Mondly
developers are operating on the belief we will be able to interact with assistants as intelligent
as Joi from Blade Runner 2049. While this may seem like a faraway dream, Joi is nothing
more than chatbot technology with speech recognition combined with mixed reality and
artificial intelligence. Mondly has already created something not too far from this fictional
scenario. Mondly’s language learning assistant from the AR module uses augmented reality
and a chatbot with speech recognition that has the ability to do things like make planets,
animals and musical instruments magically appear into a learner’s own environment, creating
an environment where people can walk up and around these virtual creations and even interact
with them. This spatial interaction creates a truly immersive one-to-one experience and is the
future of language learning for Mondly.
Currently, the chatbot experience in Mondly’s AR module aims to be the closest thing to a real-
life interaction. Mondly’s chatbot understands spoken language, replies with a human voice,
changes outfits to match the topic of discussion, and uses gestures and facial expressions to
create dynamic dialogues. All in all, the ultimate goal of chatbot technology is to be as real as
possible—to keep memories, think, and speak exactly as a human being—and replicate neural
networks that resemble the human nervous system. Mondly is striving to create a chatbot that is
our confidant and friend, one that can help and teach us whatever we need, whenever we need it,
and to make us feel emotionally connected to it.
Steps Forward
Both of the chatbots discussed here have the propensity to be useful language partners: Cleverbot
as a standalone conversant and Mondly’s chatbot as part of a broader platform for language
learning. Despite their budding usefulness, neither of them—much like chatbots more
generally—have yet to make a substantial impact.
As noted, this is in part due to chatbots’ still-developing skills for sustained conversation. Given
many users’ willingness to chat with bots (Hill, Ford, & Farreras, 2015), their potential
motivational benefits (Fryer & Carpenter, 2006; Fryer et al., 2019), their budding skills for
engaging exchanges (Coniam, 2008, 2014; Fryer & Nakao, 2009; Fryer, Nakao & Thompson,
2019), and increasing opportunities to engage users visually via AR/VR, it is the position of this
review that there are as yet still untapped opportunities for developers to make the best of the
technology we currently have. Building on the stated directions of the Cleverbot and Mondly
developers, the remainder of the discussion will focus on how current chatbots might be
structured to make the technology more useful to language learners and meaningfully start us
down a road toward the golden age in language learning to which we (i.e., researchers and
developers) aspire.
Learning in Digital Environments
Prior to discussing specific strategies for how currently available chatbot technology might be
used, a brief highlight first needs to be provided regarding what is known about learning from
digital media and agents to ensure that a sound foundation is being set. This field has rapidly
grown over the past two decades, maturing into a valuable, but often under-utilised, resource for
developers, educators and students. Mayer and colleagues’ extensive programme of research in
14
Language Learning & Technology
this area has perhaps made the single largest contribution, with Mayer’s extensive set of outputs
broadly summarising the state of the field. Many of the findings presented in Mayer’s reviews
(e.g., Mayer, 2017) are directly relevant to the development and effective use of chatbots as tools
for language learning.
The majority of the findings from Mayer’s reviews relate to supporting learners’ cognitive
processing of media delivery via digital means. Cognitive processing refers to how learners
process and thereby learn new things in these environments—how these processes are interfered
with. This body of work has established that people learn better from words and pictures than
words alone. Supporting learning in employing dual channels of learning can be a strong support
for acquiring new concepts and this might also support language acquisition and fluency
development. As outlined below, Mayer (2017) presents five effective means of reducing
extraneous processing (not overloading the user with extraneous information that does not
support the learning goal) and three means of supporting learners in learning with multimedia
information:
Reducing extraneous processing:
1. Coherence: Individuals learn better when extraneous material is excluded
2. Signalling: Individuals learn better when material is highlighted
3. Redundancy: Individuals learn better from graphics and narration than from graphics,
narration, and text
4. Spatial contiguity: Individuals learn better when corresponding words and graphics are
connected
5. Temporal contiguity: Individuals learn better when corresponding narration and graphics
are presented at the same time
Managing essential processing
1. Segmenting: Individuals learn better when a multimedia lesson is presented in small user-
paced segments
2. Pre-training: Individuals learn better when they learn key terms prior to receiving a
multimedia lesson
3. Modality: Individuals learn better from multimedia lessons when words are presented in
spoken form
The reader is referred to Mayer, 2017 and Clark & Mayer, 2016 for greater detail and examples
of applications in this growing field. Mayer and colleagues (see Mayer, 2017 for an overview)
also established three specific guidelines that are directly relevant to chatbot development and
implementation. Mayer refers to these as fostering generative processing in e-learning:
1. Individuals learn better when the words are presented in a conversational style
2. Individuals learn better from a human voice than a machine voice
3. Individuals learn better when an onscreen agent uses humanlike gestures and movement
Some of Mayer and colleagues’ recommendations will be revisited in the section below, which
focuses on directions for enhancing the usefulness of current chatbot technology. In the context
of broad principles for supporting e-learning, the key factors that Mayer and colleagues isolate
would, if addressed, support developers in both reducing extraneous processing as well as
supporting learners in managing essential processing. All of these strategies have been rigorously
tested, often in both natural and experimental settings, and have been found to substantially
Luke K. Fryer, David Coniam, Rollo Carpenter, and Diana Lăpușneanu.
15
enhance learning outcomes (Mayer, 2017). Present and future chatbot developers should
consider addressing these multimedia learning related issues, particularly for chatbots with
educational aims.
Adjusting How Chatbots Are Used and Organised
Mayer’s Principles for Generative E-learning
The first guideline (i.e., Individuals learn better when the words are presented in a conversational
style) is common sense for developers seeking to create a natural conversant, but it also suggests
that chatbots whose development is powered by input from learners might have the upper hand
over those that are programmed more explicitly. It seems reasonable to assume that there is a
middle ground. Mayer’s finding that users learn more when interacting with a natural voice lends
support to developers creating their own voice synthesis rather than relying on those built into
platforms such as Android and iOS: Mondly’s chatbot is a good example of positive initial
development in this area. Natural (i.e., consistent with human interaction) can be taken further
than simply sounding natural, as it might be important for chatbots to have distinctive voices that
suit their age and character. This is an unexplored area of development, but one that has enough
general evidence to support further innovation and testing. Mayer and colleagues’ final
suggestion— the importance of human-like chatbots—is key to supporting generative e-learning
experiences: The chatbot should utilise non-verbal means of communication, such as gestures
and movements. This indicates that animated chatbots (with arms) should be developed and
improved alongside the verbal communication software. This is an area in which Mondly’s
VR/AR ambitions might excel given time, but one that any chatbot developer might include by
integrating additional animation into its current avatars.
Multiple Chatbots
Commentators on the current communicative ability of chatbots regularly agree that chatbots
have not graduated to coherent communication beyond a few exchanges (Danilava, Busemann,
Schommer, & Ziegler, 2013; Höhn, 2017; Knight, 2017), and often only to a single exchange
of question-answer. When the chatbot fails to reply clearly or continue along a line of thought,
user interest can quickly drop off (Fryer et al., 2019). The obvious solution to this issue (i.e.,
more robust chatbots) is certainly coming, but until it does, using multiple chatbots
simultaneously might fill the gap (Candello, Vasconcelos, & Pinhanez, 2017). Multiple
chatbots, each providing different answers, and asking different questions might help learners
get enough input to ensure that they persist (which is essential for language learning). One key
to making such an approach successful is utilising the right balance of chatbot personalities
and working out a clear mechanism for interacting with the group successfully. This is one
straightforward addition to how current technology might be used more effectively right now.
Chatbots for Specific Audiences
At the moment, most chatbots are designed to communicate with more or less any kind of user.
While current chatbot technology might not be sufficient to develop these kinds of broad,
seamless communicators, narrowing the focus to specific kinds of learners could substantially
strengthen their usefulness. It is an area that language learning applications like Duolingo and,
to a greater degree Mondly, are already exploiting.
16
Language Learning & Technology
For standalone chatbots not part of language platforms, a simple option to enable the chatbot to
focus on linear questioning might be of direct benefit to some learners. While simple question-
answer interaction might bore adults, it could be a useful tool for children, both learning a new
language and developing their native language (Tewari, Brown, & Canny, 2013).
A second line of development has already been explored (Nguyen, Morales, & Chin, 2017),
that is, celebrity chatbots. A major issue with educational technology in general (Chen et al.,
2016) and language learning situations specifically (Fryer et al., 2017; 2019), is catching and
holding learners’ interest. The danger of novelty effects and language learners losing interest
quickly is a serious issue. If learners can learn about and even be able to imagine they are
talking to their favourite celebrity, they may well be more likely to get the language practice
they need (Nguyen et al., 2017). Related to this line of chatbot incentives are chatbots with an
extensive knowledge of very specific topics that might be important to the user. These topics
might be specific sports, countries, movie genres, health or cars. Learners are more likely to
forgive a chatbot’s weaknesses if they are talking about (and perhaps learning about) topics
that they themselves are interested in.
Finally, two important areas that research on chatbots has identified as particularly powerful
are language skills confidence (Fryer & Carpenter, 2006) and perceptions of chatbots as an
opportunity to practice aspects of language that classroom human partners cannot or will not
engage in (Fryer et al., 2019). The preference many language learners have for practising with
a chatbot instead of a human partner has its source in the fear of making mistakes and
appearing less than competent. The benefits conferred by the fact that chatbots are not humans
is supported by early empirical research with language learners (Fryer & Carpenter, 2006) and
more recent representative surveys of chatbot users on a broad array of platforms (Brandtzaeg
& Følstad, 2017). The second, related area of chatbots being a strong language-learning partner
is that the chatbot is willing to participate in endless practice, giving learners the chance to try
out new language and solidify newly-acquired vocabulary and grammar. Another positive
difference that even most current chatbots bring to language learning situations, which many
classroom learning human partners do not, is a wide variety of language. Classroom
communication practice is a powerful and important tool, but limited by the fact that generally
all of the students are at a similar level and therefore have little new language to contribute to
the learning process of the partner (Haines, 1995). Learners facing the communication issues
that inevitably arise when seeking to practice their L2 with many current chatbots are far more
likely to persist if they see the chatbot as a unique opportunity to learn things they could not
learn otherwise (Fryer et al., 2019). Both of these implicit chatbot strengths might be drawn
upon, even accentuated, in specific chatbots aimed at supporting learners who lack confidence
or need an opportunity to practice more than a classroom partner can or is willing to do.
Similarly, specific chatbots as tools for expanding learners’ vocabulary and chunked language
is another very specific area for development in the short-term.
Chatbots’ Potential Role in Supporting Critical Components of Interactional Competence
As increasingly powerful language partners, the long-term aim of chatbots will be to support
users’ broader interactional competence (IC; Kramsch, 1986), which is an umbrella term for a
very broad set of communicative competencies. Another critical, but less often discussed, hope
for educational technology (for important exceptions see Chun, 2011), are some of the more
nuanced skills beneath this umbrella, such as pragmatics.
Luke K. Fryer, David Coniam, Rollo Carpenter, and Diana Lăpușneanu.
17
The premise for the kinds immersive experiences which might support such subtle skills (i.e.,
VILLAGE; Virtual Immersive Language Learning and Gaming Environment) has been around
for a few decades (e.g., Hamburger & Maney, 1991). Research regarding the potential role of
chatbots in these environments is more recent, but early signs are positive (Wang, Petrina, &
Feng, 2015). Chatbots can improve the user experience, making it more immersive as well as
setting the scene for the kind of interactions that might support pragmatics development. The
visual environment and how a specific chatbot looks are easy gateways to enabling the kinds of
culturally contextualized exchanges necessary for language students to progress in the fuzzier
areas of communication (i.e., beyond basic vocabulary, syntax and pronunciation).
As reviewed already, Mondly’s VR capable chatbot creates similarly immersive experiences,
but in a straightforward one-on-one manner. As chatbots like Mondly’s become more
sophisticated, they will make gestures and facial expressions a clear part of the learning
experience. The environment of the chat will be flexible, allowing simulations that make the
connections between context and speech clear. Recent evidence suggests that even simple
simulations (without chatbots) designed to enhance students’ pragmatics can be powerful
(Sydorenko, Daurio, & Thorne, 2017; Sydorenko, Smits, Evanini, & Ramanarayanan, 2019),
setting the stage for chatbots to potentially dominate this aspect of language learning in the
near future.
Preliminary Conclusions
During their first three decades, chatbots grew from exploratory software, to the broad cast of
potential friends, guides and merchants that now populate the internet. Across the two decades
that followed this initial growth, advances in text-to-speech-to-text and the growing use of
smartphones and home assistants have made chatbots a part of many users’ day-to-day lives.
Users have been trialling well-established chatbots like Cleverbot for language learning
purposes for decades. Despite research pointing to the strengths (Fryer, 2006; Hill et al., 2015;
Fryer et al., 2019) and weaknesses (Coniam, 2008; 2014; Fryer & Nakao, 2009; Fryer et al.,
2017) of casual chatbot conversation partners, scant progress towards chatbots as substantive
language learning partners has been made. In the past three years, companies like Mondly and
Duolingo have been working to fill this gap by placing chatbots at the center of their online
language learning platforms.
The golden age for language learning that chatbots promise is still on the horizon. We
(learners, educators, and developers) should not, however, wait for a paradigm shift in machine
or deep learning to herald its arrival. Alongside the efforts of current general chatbot and
online language-learning-specific developers, there are opportunities for more effective use of
what we currently have through innovative arrangements such as multiple simultaneous
chatbots (Candello et al., 2017) and celebrity chatbots (Nguyen et al., 2017). Developers might
also fruitfully collaborate with researchers in the broader area of digital multimedia learning,
which has established a number of small, but impactful measures by which media, generally,
and conversation agents, specifically, can be enhanced.
Chatbots are a new, revolutionary stage for foreign language learning. They are currently a
useful tool that continues to grow and develop. This review proposes that chatbots can be used
more effectively right now with relatively small adjustments and the application of research from
the broader field of educational technology. Developers who can work with researchers and
18
Language Learning & Technology
educators, while listening to learners needs and experiences will find fertile ground for
innovation and impact.
References
Brandtzaeg, P. B., & Følstad, A. (2017). Why people use chatbots. Paper presented at the
International Conference on Internet Science, Thessaloniki, Greece. Abstract retrieved from
https://www.researchgate.net/publication/318776998_Why_people_use_chatbots
Cameron, G., Cameron, D., Megaw, G., Bond, R., Mulvenna, M., O’Neill, S., . . .McTear, M.
(2017). Towards a chatbot for digital counselling. Paper presented at the Proceedings of the
31st British Computer Society Human Computer Interaction Conference, Sunderland, UK.
Retrieved from https://www.scienceopen.com/document_file/cf655a90-1b6b-4a66-ab9a-
9f391a983bb5/ScienceOpen/001_Cameron.pdf
Candello, H., Vasconcelos, M., & Pinhanez, C. (2017). Evaluating the conversation flow and
content quality of a multi-bot conversational system. Paper presented at the Brazilian
Symposium on Human Factors in Computing Systems, Joinville, SC, Brazil. Abstract
retrieved from
https://www.researchgate.net/publication/320929870_Evaluating_the_conversation_flow_an
d_content_quality_of_a_multi-bot_conversational_system
Carpenter, R. (2016). Analogies and Intelligence [Web Blog post]. Retrieved from
https://www.existor.com/2016/04/11/analogies-and-intelligence/
Chen, J. A., Tutwiler, M. S., Metcalf, S. J., Kamarainen, A., Grotzer, T., & Dede, C. (2016). A
multi-user virtual environment to support students’ self-efficacy and interest in science: A
latent growth model analysis. Learning and Instruction, 41, 11–22. Retrieved from
https://doi.org/10.1016/j.learninstruc.2015.09.007
Chun, D. (2011). Computer-Assisted Language Learning. In E. Hinkel (Ed). The Handbook of
Research in Second Language Teaching and Learning. Routledge: New York. (pp. 663–
680).
Clark, R. C., & Mayer, R. E. (2016). E-learning and the science of instruction: Proven
guidelines for consumers and designers of multimedia learning. Hoboken, N J: John Wiley &
Sons.
Coniam, D. (2004). Using language engineering programs to raise awareness of future CALL
potential. Computer Assisted Language Learning, 17, 149–176. Retrieved from
https://doi.org/10.1080/0958822042000334226
Coniam, D. (2008). An evaluation of chatbots as software aids to learning English as a second
language. The Eurocall Review, 13, p 1–19 Retrieved from: http://www. eurocall
languages.org/review/13/#coniam
Coniam, D. (2014). The linguistic accuracy of chatbots: Usability from an ESL perspective. Text
& Talk, 34, 545–567. https://doi.org/10.1515/text-2014-0018
Crutzen, R., Peters, G.-J. Y., Portugal, S. D., Fisser, E. M., & Grolleman, J. H. (2011). An
artificially intelligent chat agent that answers adolescents’ questions related to sex, drugs, and
alcohol: An exploratory study. Journal of Adolescent Health , 48, 514–5199. Retrieved from
https://doi.org/10.1016/j.jadohealth.2010.09.002
Luke K. Fryer, David Coniam, Rollo Carpenter, and Diana Lăpușneanu.
19
Danilava, S., Busemann, S., Schommer, C., & Ziegler, G. (2013). Towards computational
models for a long-term interaction with an artificial conversational companion. Proceedings
of the 5th International Conference on Agents an Artificial Intelligence, (pp. 241-248).
Barcelona, Spain .Retrieved from http://orbilu.uni.lu/bitstream/10993/12575/1/ACC-
ICAART2013.pdf
Daniel, F., Matera, M., Zaccaria, V., & Dell’Orto, A. (2018). Toward truly personal chatbots: On
the development of custom conversational assistants. In Proceedings of the 1st International
Workshop on Software Engineering for Cognitive Services, SE4COG ’18, pages 31– 36, New
York, NY, USA, 2018. ACM. Retrieved from
https://re.public.polimi.it/bitstream/11311/1071751/1/bot.pdf
Dale, R. (2016). The return of the chatbots. Natural Language Engineering, 22, 811–817.
Retrieved from https://doi.org/10.1017/S1351324916000243
Dubosson, F., Schaer, R., Savioz, R., & Schumacher, M. J. S. M. I. (2017). Going beyond the
relapse peak on social network smoking cessation programmes: ChatBot opportunities. Swiss
Medical Informatics. 33, 1–6. Retrieved from
https://www.researchgate.net/publication/324904852_Going_beyond_the_relapse_peak_on_s
ocial_network_smoking_cessation_programmes_ChatBot_opportunities
Fleming, M., Riveros, P., Reidsema, C., & Achilles, N. (2018). Streamlining student course
requests using chatbot. In 29th Australasian Association for Engineering Education
Conference 2018 (AAEE 2018). Hamilton, New Zealand: Engineers Australia, 2018: 207-
211. Retrieved from
https://search.informit.com.au/documentSummary;dn=166919861347542;res=IELENG
Fryer, L. K. (2006). Bots for language learning. The Language Teacher, 30, 33–34. Retrieved
from http://jalt-publications.org/tlt/issues/2006-08_30.8
Fryer, L. K., & Carpenter, R. (2006). Bots as language learning tools. Language Learning &
Technology, 10, 8–14. Retrieved from http://llt.msu.edu/vol10num3/emerging/default.html
Fryer, L., & Nakao, K. (2009). Assessing chatbots for EFL learner use. pp 849-857 In A. Stoke
(Ed.), JALT2008 Conference Proceedings. Tokyo: JALT. Retrieved from:
http://jaltpublications.org/proceedings/articles/84-jalt2009-proceedings-contents
Fryer, L. K., Ainley, M., Thompson, A., Gibson, A., & Sherlock, Z. (2017). Stimulating and
sustaining interest in a language course: An experimental comparison of Chatbot and Human
task partners. Computers in Human Behavior, 75, 461–468. Retrieved from
https://doi.org/10.1016/j.chb.2017.05.045
Fryer, L. K., Nakao, K., & Thompson, A. (2019). Chatbot learning partners: Connecting learning
experiences, interest and competence. Computers in Human Behavior, 93, 279–289.
Retrieved from https://doi.org/10.1016/j.chb.2018.12.023
Haines, S. (1995). For and against pairwork. Modern English Teacher, 4, 55–58.
Hamburger, H. & Maney, T. (1991). Twofold continuity in immersive language learning. CALL,
4 81–92. Retrieved from https://doi.org/10.1080/0958822910040203
20
Language Learning & Technology
Hill, J., Ford, W. R., & Farreras, I. G. (2015). Real conversations with artificial intelligence: A
comparison between human-human online conversations and human-chatbot conversations.
Computers in Human Behavior, 49, 245–250. Retrieved from
https://doi.org/10.1016/j.chb.2015.02.026
Hirschberg, J., & Manning, C. D. (2015). Advances in natural language processing. Science, 349,
261–266. Retrieved from https://doi.org/10.1126/science.aaa8685
Höhn, S. (2017). A data-driven model of explanations for a chatbot that helps to practice
conversation in a foreign language. Paper presented at the Proceedings of the 18th Annual
SIGdial Meeting on Discourse and Dialogue, Saarbrücken, GermanyRetrieved from
https://www.researchgate.net/publication/322582675_A_data_driven_model_of_explanation
s_for_a_chatbot_that_helps_to_practice_conversation_in_a_foreign_language
Howell, D. (2019, June, 5). How voice recognition can be a major asset for your business.
Retrieved from https://www.techradar.com/uk/news/software/applications/how-voice-
recognition-can-be-a-major-asset-for-your-business-1321534
Vogels, E. A. (2018,Sept, 9). Millennials stand out for their technology use, but older
generations also embrace digital life. Retrieved from https://www.pewresearch.org/fact-
tank/2019/09/09/us-generations-technology-use/
Knight, W. (2017, June, 31). To build a smarter chatbot, first teach it a second language. MIT
Technology Review. Retrieved from https://www.technologyreview.com/s/608382/to-build-
a-smarter-chatbot-firstteach-it-a-second-language/
Kramsch, C. (1986). From language proficiency to interactional competence. The Modern
Language Journal, 70, 366–372. Retrieved from https://doi.org/10.1111/j.1540-
4781.1986.tb05291.x
Kriegeskorte, N. (2015). Deep neural networks: A new framework for modelling biological
vision and brain information processing. Annual Review of Vision Science, 1, 417–446.
Retrieved from https://www.biorxiv.org/content/biorxiv/early/2015/10/26/029876.full.pdf
Mayer, R. E. (2017). Using multimedia for e-learning. Journal of Computer Assisted Learning,
33, 403–423. Retrieved from https://doi.org/110.1111/jcal.12197
Meszaros E.L., Chandarana M., Trujillo A., Allen B.D. (2018) Compensating for Limitations in
Speech-Based Natural Language Processing with Multimodal Interfaces in UAV Operation.
In: Chen J. (eds) Advances in Human Factors in Robots and Unmanned Systems. AHFE
2017. Advances in Intelligent Systems and Computing, vol 595. Springer, Cham. Retrieved
from https://link.springer.com/chapter/10.1007%2F978-3-319-60384-1_18
Nguyen, H., Morales, D., & Chin, T. (2017). A Neural Chatbot with Personality [unpublished
manuscript]. Retrieved from
https://web.stanford.edu/class/archive/cs/cs224n/cs224n.1174/reports/2761115.pdf
Shah D., Philip T.J. (2019) An Assistive Bot for Healthcare Using Deep Learning:
Conversation-as-a-Service. In: Pati B., Panigrahi C., Misra S., Pujari A., Bakshi S. (Eds)
Progress in Advanced Computing and Intelligent Engineering. Advances in Intelligent
Systems and Computing, vol 713. Springer, Singapore
Luke K. Fryer, David Coniam, Rollo Carpenter, and Diana Lăpușneanu.
21
Simonite, T. (2017, April 17). Facebook’s perfect impossible chatbot. MIT Technology Review.
Retrieved from https://www.technologyreview.com/s/604117/facebooksperfect-impossible-
chatbot/
Smith PK, Pellegrini A. Learning Through Play. In: R. E. Tremblay, M. Boivin, & R. Peters
(Eds.); P. K. Smith (Topic ed). Encyclopedia on Early Childhood Development. Retrieved.
http://www.child-encyclopedia.com/play/according-experts/learning-through-play
Sydorenko, T., Daurio, P., & Thorne, S. L. (2017). Refining pragmatically appropriate oral
communication via computer-simulated conversations. Computer Assisted Language
Learning, 31, 1–25.Retrieved from https://doi.org/10.1080/09588221.2017.1394326
Sydorenko, T., Smits, T. F. H., Evanini, K., & Ramanarayanan, V. (2019). Simulated speaking
environments for language learning: Insights from three cases. Computer Assisted Language
Learning, 32, 17-48. Retrieved from https://doi.org/110.1080/09588221.2018.1466811
Tewari, A., Brown, T., & Canny, J. (2013). A Question-answering Agent Using Speech Driven
Non-linear Machinima. Berlin, Heidelberg: Springer Berlin Heidelberg.
Vogels, A. (2019, Sept, 9). Millennials stand out for their technology use, but older generations
also embrace digital life. Facttank. Downloaded from https://www.pewresearch.org/fact-
tank/2019/09/09/us-generations-technology-use/ on March 6, 2020.
Weizenbaum, J. (1966). ELIZA—A computer program for the study of natural language
communication between man and machine. Communications of the ACM, 9(1). Retrieved
from http://i5.nyu.edu/~mm64/x52.9265/january1966.htm
Wang, Y. F., Petrina, S., & Feng, F. (2015). VILLAGE-Virtual Immersive Language Learning
and Gaming Environment: Immersion and presence. British Journal of Educational
Technology, 48, 431–450. Retrieved from
Zhou, L., Gao, J., Li, D., & Shum, H.-Y. (2018). The Design and Implementation of XiaoIce, an
Empathetic Social Chatbot [unpublished manuscript]. Retrieved from
https://arxiv.org/pdf/1812.08989
About the Authors
Luke K. Fryer is an Associate Professor, Faculty of Education (CETL), HKU. His research
addresses motivations and strategies for learning on and offline. A considerable portion of his
recent work is focused on interest development within and across formal education. He has a
longstanding, personal interest in the potential role of Bots within student learning.
E-mail: lukefryer@yahoo.com
David Coniam is Research Chair Professor and Head of Department of the Department of
Curriculum and Instruction in the Faculty of Education and Human Development at The
Education University of Hong Kong, where he is a teacher educator, working with teachers in
Hong Kong primary and secondary schools. His main publication and research interests are in
language assessment, language teaching methodology and computer assisted language learning.
E-mail: coniam@eduhk.hk
Rollo Carpenter is the creator of Cleverbot, AI that engages millions of people around the world
in conversation. Cleverbot started years earlier than bots like Siri and Alexa, and talks not to
22
Language Learning & Technology
assist but to entertain. British born, Rollo has an MA from Oxford University, and to 2003 was
founder and CTO of a startup in California
E-mail: rollo@existor.com
Diana Lăpușneanu is a Content Writer at Mondly. She has a bachelor's degree in Advertising at
the University of Bucharest where she created a paternal leave promotional campaign for
Romanian dads; and a master's degree in Image Campaign Management. Her greatest passions
are film, creative writing, and classical mythology.
E-mail: diana@mondly.com
